Hypothalamic-Pituitary-Adrenal Axis Attenuation Following Child Sexual Abuse Distinguishes Posttraumatic Stress Trajectories from Late Adolescence to Midlife

Abstract

Biological embedding describes the process of responsive developmental adaptation whereby stress “gets under the skin.” Applied to the endocrine system, early chronic stress exposure may contribute to hypothalamic–pituitary–adrenal (HPA) axis attenuation, characterized by initial elevations in cortisol followed by less steep normative increases across development due to cumulative allostatic load. Simultaneously, posttraumatic stress symptoms (PTSS), including hyperarousal and hypervigilance, may potentially reinforce HPA attenuation in a feedback loop. Prospective, longitudinal assessments of HPA attenuation are rarely undertaken, obfuscating physiological underpinnings of posttraumatic stress. Using data from 164 female participants (45.6% with confirmed child sexual abuse exposure), we approximated HPA attenuation by estimating morning cortisol intercepts and slopes from early adolescence to early adulthood. Growth mixture modeling was then used to identify distinct longitudinal patterns of PTSS from late adolescence to midlife. Four trajectories were identified: persistently high symptoms across adulthood (High Stable), increasing symptoms emerging over time (Increasing), declining symptoms following initially elevated levels (Decreasing), and consistently low symptoms (Low Stable). Cortisol intercepts and slopes were both significantly associated with PTSS trajectory membership, such that greater HPA attenuation was linked with chronic symptoms. For example, each 1-standard deviation decrease in cortisol slope (reflecting greater HPA attenuation) was associated with nearly threefold higher odds of belonging to the High Stable PTSS trajectory relative to the Low Stable trajectory (OR = 2.97, 95% CI [1.40, 6.31], p = .005). Findings suggest etiologically distinct pathways to chronic and delayed posttraumatic stress, underscoring the importance of biologically-informed, developmentally tailored intervention strategies.

Graphical Abstract

1.0. Introduction

Child sexual abuse (CSA) is pervasive, affecting 1 in 4 women in the United States.¹ The high prevalence of CSA raises concern, as CSA is linked with myriad negative developmental consequences.² Relevant to the current investigation, CSA is associated with elevated lifetime posttraumatic stress disorder (PTSD) diagnosis across studies assessing PTSD generally as well as PTSD specifically linked to CSA.³ Clarifying potential mechanisms, particularly biological mechanisms, that may explain how CSA shapes the course of posttraumatic stress symptoms (PTSS) over time will help elucidate the biological underpinnings of PTSD. The current study prospectively evaluated how physiological indicators of biological embedding, as indexed by hypothalamic-pituitary-adrenal (HPA) axis attenuation over development, may predict differentiated patterns of PTSS from late adolescence to mid-life among women with and without confirmed CSA-exposure.

The HPA-axis is critical to orchestrating physiological stress responses to environmental threat across development. Activation of the HPA-axis frees up metabolic resources, inhibits sympathetic and adrenomedullary systems to down-regulate the fight or flight response, and promotes behavioral adaptation through consolidation of emotionally significant events. Notably, prior research suggests HPA function is not static over development. Rather, cortisol output changes systematically with age, pubertal maturation, and accumulated environmental demands.^4–6 Despite its central role in the stress response, prior research indicates associations between HPA axis function and PTSD diagnosis are somewhat equivocal,^7,8 and may depend on timing of exposure and method of assessing HPA-axis functioning.⁹ These inconsistencies suggest that static, cross-sectional measures of cortisol may obscure meaningful developmental processes through which stress physiology becomes embedded and shapes later mental health risk.

Biological embedding frameworks^10,11 provide a developmentally grounded account of how early chronic stressors can become instantiated in biological systems in ways that shape subsequent adaptation. Biological embedding frameworks further propose that early chronic stress acts on several biological systems, including epigenetic modifications, neurodevelopment, and relevant to the current study, physiological alterations.¹¹ These frameworks further posit that stress response systems recalibrate over time in response to repeated demand, maintaining short-term adaptation at the potential cost of long-term physiological wear and tear (i.e., allostatic load).¹² Within the HPA-axis, this recalibration may be reflected in progressive attenuation in cortisol levels across development as sustained activation gives way to downregulation at peripheral levels of the axis.¹³ Applied to the current study, we argue biological embedding theories may point to attenuated developmental trajectories of morning cortisol as a potential stress adaptation mechanism with implications for other areas of functioning. From this perspective, HPA attenuation represents a developmental process that unfolds gradually as individuals adapt to ongoing stress exposure rather than an abrupt failure.^14,15 We argue this formulation that emphasizes developmental HPA attenuation can help resolve longstanding inconsistencies in the adversity literature^16–23 by shifting away from evaluating whether cortisol is “low” or “high” at a given age and instead characterizing how cortisol levels change across sensitive developmental periods.

To this end, the current study follows up on a landmark paper by Trickett and colleagues¹³ conducted with the same cohort. Trickett examined linear trends in morning basal cortisol from age 6 to 30, finding that CSA-exposed participants had significantly higher morning cortisol in the period after initial disclosure of CSA, as indexed by cortisol intercepts at baseline. However, there was a significant CSA by cortisol slope interaction, such that CSA-exposed participants exhibited a significantly less steep linear slope of morning cortisol over development. These results provide support for the HPA attenuation hypothesis,²⁴ whereby early adversity may result in heightened HPA-axis activity initially, but over the course of development, the cumulative and unsustainable demands of this HPA activity ultimately leads to blunted cortisol levels in adulthood. The current study builds on prior findings with this cohort to investigate how HPA attenuation may be implicated in the course of PTSS. While prior work has separately linked (a) CSA with PTSS and (b) cortisol functioning with PTSS, suggesting HPA functioning may serve as an intermediate mechanism, most existing studies have relied on cross-sectional designs that are not well suited to assessing long-term HPA attenuation over development. While this operationalization of HPA attenuation used in both the Trickett study and the current investigation provides a novel developmental lens, it should be noted that this method relies on morning basal cortisol data as the primary index of HPA function. Morning cortisol reflects a daily peak activation or “set-point” and should be distinguished from other HPA indices such as diurnal cortisol slope or area under the curve (AUC), which reflect cumulative daily cortisol output.

Further supporting the HPA attenuation hypothesis tested in earlier work with this cohort, a meta-analysis by Miller and colleagues¹⁴ on effects of chronic stressors on HPA axis function in humans found that the time elapsed since the onset of the stressor moderated the association between stress and HPA function, such that effect sizes decreased as time since the onset of stress increased (morning cortisol samples: b = −.13, p < .01; all types of cortisol samples: b = −.29, p < .01). Interestingly, findings from this meta-analysis¹⁴ also indicated that among people who developed PTSD after chronic stressors, total daily cortisol output was lower (d = −.32, p < .01). Although the current investigation focuses on morning cortisol rather than total output, this prior finding suggests that the long-term HPA attenuation pattern observed after chronic stress may be compounded by PTSD symptomology in a potential dynamic, reinforcing feedback loop. This is consistent with the emphasis of biological embedding frameworks on dynamic, potentially bidirectional transactions between stress biology and psychological processes over time.^25,26 For example, HPA attenuation may alter emotional and cognitive responses to subsequent stressors, and in turn, persistent posttraumatic stress symptoms like hypervigilance, intrusive recollections, and altered threat appraisal may further tax stress response systems, reinforcing patterns of HPA attenuation.

Drawing on a prospective, 30-year study of an all-female sample with rigorous demographic matching procedures designed to isolate effects of CSA on development, the current study aims to evaluate developmental HPA attenuation as an etiological mechanism that may contribute to variation in PTSS symptom course following CSA exposure. Underscoring the strength of the current sample for addressing this question, prior meta-analyses indicate exposure to CSA is associated with over double the risk for developing PTSD in women (OR = 2.38).²⁷ Thus, CSA confers higher estimated risk for PTSD than several other traumatic events, including sudden bereavement, traffic accidents, and war.²⁸ Moreover, women develop PTSD at higher rates than men, even after accounting for the type of potentially traumatic event that precipitated development of PTSD.²⁹ Much of the foundational research on PTSD was conducted with combat-exposed men, and while recent research has endeavored to incorporate more focus on women’s experiences of PTSD, women remain comparatively underrepresented in PTSD research overall.³⁰ Furthermore, elevated rates of CSA and PTSD in women may help explain some of the observed health disparities in women, such as higher rates of disabilities, chronic pain, and autoimmune diseases.³¹ Thus, the current study is well-poised to inform public health by clarifying posttraumatic stress etiology and course in populations that bear a disproportionate share of trauma exposure and downstream health burden.

Furthermore, clinical theories and conceptualizations of PTSD, including Bonanno’s Stress-Response and Resilience Framework³² emphasize normal variation in adaptive functioning following trauma exposure. Bonanno’s framework also posits resilience, defined as low stable symptoms over time, as the modal trauma response,^33,34 highlighting the importance of studying both risk and resilience. Importantly, these hypothesized variations in PTSS trajectory over time represent clinically meaningful patterns that may correspond to different therapeutic needs, mechanisms of change and etiology, and levels of risk and resilience. Understanding the factors that shape divergence into these courses may ultimately support more personalized intervention planning and underscores the practical value of continued investigation into their antecedents.

To this end, the current study uses a continuous measure of PTSS, capturing developmental trends and subclinical symptoms. Examining trajectories is preferable to binary PTSD diagnostic categories, as many individuals experience fluctuating symptoms, worsening symptoms, and even spontaneous recovery. Identifying individuals at risk for chronic PTSS is an important public health aim, as chronic posttraumatic stress may incur neurobiological damage including reduced hippocampal volume and hyperthyroidism.³⁵ To capture variation in the unfolding of PTSS symptoms over time and development, we applied person-centered statistical approaches, which are well-suited to capture individual typologies that may be obscured by the focus on aggregate effects in variable-centered approaches. More specifically, latent growth mixture modeling (GMM) applies person-centered approaches longitudinally, thus parsing heterogeneity in individual change over time by identifying data-driven, prototypical trajectories of development.³⁶ Prior comprehensive reviews³⁷ have identified four modal observed trajectories following trauma exposure: recovery, chronic, resilience, and delayed onset. These four modal trajectories have also been identified in prior studies of very similar samples³⁸ (i.e., all female, CSA-enriched samples), albeit on a shorter timescale. In summary, by chronicling the course of PTSS from late adolescence to midlife following exposure to CSA, the current study provided a developmental lens for characterizing how CSA may shape subsequent trajectories of PTSS in adulthood via biological embedding, as indexed by HPA attenuation, with substantial clinical implications.

1.1. Current Study

This study leverages data from a longitudinal, cross-sequential study of female participants to evaluate the role of HPA attenuation in shaping developmental PTSS trajectories from late adolescence to midlife. Approximately half of participants in this cohort were exposed to substantiated CSA. Prior research with this cohort has established that CSA exposure is linked with HPA attenuation, but the role of CSA-related HPA attenuation in long range posttraumatic stress symptom course has not been characterized. To our knowledge, only two prior studies^39,40 have examined PTSS trajectories over such a long period of time, both of which were based on samples of military veterans. However, based on previous research,^37,38 we hypothesized the emergence of four PTSS trajectories, corresponding to chronically high PTSS, resilience as indicated by low stable PTSS, decreasing PTSS indicative of recovery, and late onset PTSS characterized by increasing PTSS over time. We further hypothesized that resilience (i.e., low stable PTSS) would be the modal trajectory. We also hypothesized that CSA-exposure would be associated with symptomatic PTSS trajectories, particularly chronically high PTSS trajectories, and that subsequent revictimization would increase likelihood of belonging to the late symptom onset (increasing PTSS) trajectory. Most centrally, we hypothesized that biological embedding would be implicated in PTSS trajectories. Specifically, we expected biological embedding in the HPA-axis (i.e., HPA attenuation as indicated by higher initial morning cortisol levels and less steep cortisol slopes over time) would be associated with membership in PTSS trajectories characterized by chronic symptoms.

2.0. Methods

2.1. Participants & Procedures

The current cohort includes 187 females recruited for a larger, longitudinal, cross-sequential design study⁴¹ of the effects of CSA on development, of which 164 were included in the present analysis. Eligibility criteria included (a) age 6–16 years old at baseline, (b) participation within 6 months of disclosure of abuse to authorities, (c) substantiated sexual abuse including genital contact and/or penetration, (d) perpetration by a family member, and (e) participation in the study of a non-abusing caregiver (usually the biological mother). However, 16 participants from the demographically matched comparison group were later identified as having non-criteria abuse (i.e., unsubstantiated sexual abuse, sexual abuse without genital contact and/or penetration, non-family perpetrator), and were thus excluded. An additional seven participants were missing PTSS data at timepoints T4, T6, and T7, resulting in a final sample of 164 for the present analyses. Of the 164 participants analyzed here, 78 (45.6%) experienced CSA. Sexually abused participants were referred to the study by Child Protective Services (CPS). Comparison participants were recruited via adverts in newspapers and community organizations in the same neighborhoods from which CSA-exposed participants were recruited. Comparison and CSA-exposed participants were recruited to be demographically similar across racial/ethnic group, age, socioeconomic status, family constellation, nonsexual trauma exposure, and residing zip code. Data are not publicly available due to the sensitive nature of the study.

Participants’ age at baseline ranged from 6 to 16 years old (M_age = 11.1 years). Participants were followed over 7 waves of data collection from 1987–2019 (see Table 1 for more details) and were on average 36.8 years old (SD = 3.64) at T7. Regarding retention, 98.4% of participants completed at least one follow-up visit and 96.8% completed 2 or more follow up visits. The median number of timepoints completed was 6 (out of 7 total possible timepoints). 53.5% of participants identified as White, 42.9% identified as Black or African American, 2.9% identified as Hispanic, and 0.6% identified as Asian, see Table 1. The mean Hollingshead score for participants’ family of origin was 37.20 (SD = 11.19), indicating participants were on average working to middle class.

Table 1.

Sample characteristics & longitudinal design

1a. Sample characteristics
Variable	Range	Mean (SD)
Hollingshead Score	12–63	37.20 (11.19)
# Non-Sexual Traumatic Events	0–21	8.11 (5.01)
T4 Total PTSS	0–17	7.72 (4.93)
T4 Re-experiencing Subscore	0–5	2.45 (1.59)
T4 Avoidance Subscore	0–7	2.61 (1.91)
T4 Arousal Subscore	0–6	2.66 (2.10)
T6 Total PTSS	0–18	6.54 (5.14)
T6 Re-experiencing Subscore	0–5	2.09 (1.59)
T6 Avoidance Subscore	0–7	2.26 (2.14)
T6 Arousal Subscore	0–6	2.20 (1.99)
T7 Total PTSS	0–18	9.42 (5.43)
T7 Re-experiencing Subscore	0–5	2.81 (1.71)
T7 Avoidance Subscore	0–6	3.40 (2.30)
T7 Arousal Subscore	0–7	3.25 (2.14)
Variable	n (%)
Race/Ethnicity
White	91 (53.5%)
Black/African American	73 (42.9%)
Hispanic	5 (2.9%)
Asian	1 (.6%)
CSA status
CSA-Exposed	78 (45.6%)
No CSA	93 (54.4%)
Adult Sexual Victimization
Yes	54 (33.7)
No	106 (66.3%)

1b. Longitudinal Design & Timing of Assessments
Timepoint	Data Collection Years	Participant Age (Mean, SD in years)	CSA	Cortisol	PTSS
T1	1987–1992	11.09 (3.00)	✓	✓
T2	1988–1994	12.28 (3.01)		✓
T3	1989–1995	13.48 (3.03)		✓
T4	1996–1998	18.03 (3.45)		✓	✓
T5	1998–2000	20.32 (3.43)		✓
T6	2002–2004	24.44 (3.20)		✓	✓
T7	2014–2019	36.82 (3.64)			✓

Note. CSA = child sexual abuse. Nonsexual traumatic events refers to the total number of types of non-sexual potentially traumatic events retrospectively endorsed on the Comprehensive Trauma Interview.

This study was approved by the Institutional Review Board at [BLINDED] and a Federal Certificate of Confidentiality was obtained. Non-abusing caregivers provided consent for participants under age 18, with participants age 6–17 providing assent. After turning 18, participants consented themselves. Data collections were conducted in the morning (beginning between 8:30 and 9:00am). After completing the informed consent process, participants were instructed to relax and completed demographic questionnaire forms for 20 to 30 minutes. This period of relative calm minimized potential stress effects prior to nonstress morning cortisol sample collections. After morning cortisol samples were collected, participants completed additional questionnaires.

2.2. Measures

2.2.1. CSA Status.

CSA status was determined at study enrollment from CPS records and further confirmed via trauma interviews. In the course of trauma interviews, 16 participants were later identified as having non-criteria abuse (i.e., unsubstantiated sexual abuse, sexual abuse without genital contact, non-family perpetrator) and were thus excluded. The median age at onset of abuse was 7.5 years and the median duration was approximately 2 years. 70% of participants reported vaginal and/or anal penetration and 60% of perpetrators were a paternal figure (biological father, stepfather, or mother’s live-in boyfriend). CSA status is coded as a binary variable in analyses (0 = non-abused comparison, 1 = CSA-exposed).

2.2.2. PTSS and Trauma Exposure.

Participants completed the Comprehensive Trauma Interview (CTI)⁴² to assess PTSS across symptom groups of re-experiencing, avoidance, and arousal, as well as exposure to a host of potentially traumatic events. The CTI has been validated for use with adolescents and adults.⁴³ A score of 7 or higher indicates clinically elevated PTSS. We constructed person-centered trajectories of PTSS using total symptom scores at T4 (late adolescence, M_age = 18.0, SD = 3.4), T6 (early adulthood, M_age = 24.4, SD = 3.2), and T7 (adulthood M_age = 36.8, SD = 3.6). At T7, participants also answered 24 items retrospectively assessing lifetime experiences of other potentially traumatic events (PTEs) such as physical abuse, neglect, assault, and severe illness. The 22 non-sexual violence related items were used to create a lifetime exposure to non-sexual trauma total score, indexing the total types of traumatic events reported by participants. Adult sexual victimization was assessed via a single item: “Since you turned 18, has anyone done something sexual to you that you didn’t want?” Participants endorsed a mean of 8.1 (SD = 5.0) types of non-sexual trauma and 34% endorsed adult sexual victimization. See Table 1 for sample-level summary statistics for PTSS and trauma exposure.

2.2.3. Morning Cortisol.

Morning cortisol data were used to index HPA attenuation over development. As this long-term longitudinal study spanned approximately 30 years, available technology and best practices for morning cortisol collection and assays changed over the course of the study. From T1-T3 (1987–1992) serum assessments were the best available sampling technique. By T4 data collections, less invasive saliva sampling techniques for cortisol became available. T4-T6 (1996–2004) unbound cortisol concentrations were thus assayed from saliva. While saliva sampling reduced participant burden, this procedural change complicates longitudinal analyses, as serum and saliva assessments of unbound cortisol have differing ranges, despite being highly correlated. To account for these methodological differences, we used a formula (Equation 1) developed by Salimetrics Laboratories (State College, PA) using a US Food and Drug Administration (FDA)- approved (FDA 510[k] #K031348, June 10, 2003) salivary cortisol enzyme immunoassay investigative device. This conversion formula has also been applied in previously published investigations¹³ from this cohort.

$$y\left(\text{serum}\mu \text{g/dl}\right)=5.177+15.123\text{x}\left(\text{saliva}\mu \text{g/dl}\right)$$

Serum cortisol samples were obtained via an indwelling catheter inserted into a forearm vein and were stored at −70°C until assayed via radioimmunoassay by Hazelton Laboratories (Vienna, VA). All samples across timepoints were collected within a 1-hour time of day range (9:00–10:00am). Serum intra- and interassay variability were 3.4% and 12.3%, respectively. Saliva cortisol samples were obtained via passive drool and stored at −70°C and assayed by Salimetrics Laboratories using a highly sensitive enzyme immunoassay using 25 ml of saliva per determination. The saliva immunoassay has a lower limit of sensitivity of 0.003 ug/dl, a standard curve range of 0.007–1.8 ug/dl, and average intra- and interassay coefficients of variation 5.10% and 8.20%, respectively. Cortisol concentrations were used to compute cortisol intercepts (i.e., initial basal morning cortisol levels at T1) and cortisol slopes over time per the data analytic plan below. All references to cortisol intercepts and slopes in the current results refer to parameters derived from morning cortisol.

2.3. Data Analytic Plan

Sample size was determined based on the aims of the larger parent study. Determining sufficient statistical power for Growth Mixture Modeling (GMM) is complex and is affected by various factors (e.g., sample size, non-normality, effect size for class separation, number of classes, the number of repeated measurements for modeling the growth trajectories, group proportion, and the actual observed growth trajectories). As presented in the results below, the present study had a sample size of 164 and three repeated measurement time points that yielded a four-class solution in the GMM. Thus, prior simulation studies^44,45 suggest that the current analyses are likely underpowered. Although the interpretation of the GMM classes was guided by theory and not solely data driven, findings should still be interpreted in light of this power limitation.

Data were screened for outliers and non-normality in SPSS 29.0. None exhibited excessive skewness or kurtosis (see tables S1, S2). A GMM was performed in Mplus 8.1. Full information maximum likelihood (FIML) estimation was used to address missing data given its superiority to listwise deletion or other ad hoc methods.⁴⁶ GMM captures heterogeneity in growth trajectories by estimating latent intercepts and slopes and testing whether two or more distinct classes of individuals exist, and further determining optimal class membership for each individual. Trajectories of PTSS scores were modeled from T4 to T6 to T7, with approximately 6 years elapsed between T4 and T6, and 12 years elapsed between T6 and T7. Thus, time coding was set @0 for T4, @1 for T6, and @3 for T7, reflecting that the time elapsed between T6 and T7 was roughly double the time elapsed between T4 and T6. The best fitting GMM solution was determined using several fit indices while also considering the theoretical interpretability and practical utility. Fit indices used included the Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and adjusted Bayesian Information Criterion (aBIC), where lower values indicate better fit, as well as the Lo-Mendell-Rubin Adjusted Likelihood Ratio Test (aLRT), which tests whether an n profile solution fits significantly better than the n-1 model.

After identifying the best-fitting class solution, we evaluated whether CSA exposure predicted PTSS trajectory membership using the R3Step Approach,⁴⁷ which uses mixture modeling to incorporate auxiliary variables into the model using a stepwise process, as follows: Step 1) Estimate standard GMM without predictors; Step 2) Assign participants to latent classes based on posterior probabilities; Step 3) Estimate a logistic regression model with assigned class membership as the dependent variable. Step 3 incorporates measurement error, accounting for uncertainty in class assignment. Prior simulation work⁴⁸ shows that under conditions of entropy of 0.6 or higher, the R3Step approach outperforms both exported class membership methods and multiple imputation methods in terms of bias and confidence interval coverage, with negligible effects of sample size.

To index individual participants’ HPA axis functioning, an unconditional multilevel model (MLM) was constructed to estimate the sample average growth trajectory of raw morning cortisol scores and individual deviations from the sample trajectory from T1 to T6, corresponding to cortisol slopes from mean age ≈ 11 to 24. Morning cortisol data did not exhibit excessive non-normality (skewness < |2.0| and kurtosis < |7.0|)^49,50 at any timepoint (see Table S1), and thus MLMs were run on raw (rather than transformed) morning cortisol data to enhance comparison to other publications^13,51 that have used these same cortisol slopes and intercepts with this cohort. Random individual intercept and the linear slope parameter estimates, representing serum basal morning cortisol level in childhood (mean age of approximately 11 (T1); “initial cortisol level”) and morning cortisol growth rate over time (“cortisol slope” from T1-T6), respectively, were extracted for use as predictors in subsequent analyses. Time was indexed in years since the baseline data collection visit, which per protocol, occurred within 6 months of report of CSA to authorities. Higher values of initial cortisol level and lower values of cortisol growth rate (i.e., less steep slopes) indicated greater HPA attenuation.¹³ The use of linear cortisol slopes (as opposed to quadratic or other curvilinear parameters) is supported by previous work with this cohort.^13,51 Specifically, although both linear and quadratic models exhibited adequate fit to this cohort’s cortisol data,¹³ only linear cortisol slopes 1) showed significant differences between CSA-exposed and non-abused comparisons¹³ and 2) were predictive of subsequent clinically relevant outcomes, such as obesity status.⁵¹

We tested the predictive effect of HPA axis attenuation (individual participants’ cortisol intercept and slope values) on PTSS trajectory membership using the R3Step approach, which yields unstandardized regression coefficients, significance levels, and odds ratios reflecting the association between HPA attenuation indicators (cortisol intercept and slope) and PTSS trajectory membership. We further supplemented the R3Step analysis with the BCH⁵² approach for descriptive purposes in order to provide the average cortisol intercept and slope membership for each PTSS trajectory, which may enhance interpretation, though these results are provided only in tables and not discussed in text. While ample guidance for incorporating covariates into R3Step and BCH models exists for latent class and profile analysis, there is little to no current guidance for incorporating covariates into these approaches in GMM models. Thus, to demonstrate that associations between HPA attenuation and PTSS trajectory were robust over and above the effects of demographic variables, we performed a supplemental analysis in which we created residualized cortisol intercept and slope values for each participant by regressing maternal education level and participant racial minority status on their cortisol intercept and slope scores and saving out the residual, thus representing participants’ HPA attenuation after controlling for demographic confounds. Maternal education and racial minority status were included as covariates because they represent social determinants linked to early-life stress exposure and health disparities, and thus represent plausible confounds in associations between cortisol attenuation and posttraumatic stress. We then reran R3Step models with residualized cortisol slope and intercept scores as the auxiliary variables, as presented in the sensitivity analysis section.

Finally, as a post-hoc analysis to further contextualize our findings, we also used the R3Step approach to test associations between lifetime trauma exposure as well as adult sexual revictimization and PTSS trajectory membership, hypothesizing that those individuals who endorse subsequent trauma and sexual revictimization would be more likely to belong to the high stable or increasing PTSS trajectories relative to other trajectories based on previous research.⁵³

3.0. Results

3.1.1. Class Enumeration

GMM solutions with 1 through 6 classes were tested, as presented in the model fit indices in Table 2. The AIC (AIC_4-class = 2251.840) and aBIC (aBIC_4-class = 2250.915) both indicated the 4-class solution had the best fit, whereas the unadjusted BIC suggested both the 2- and 4-class solution could be viable (BIC_2-class = 2279.366, BIC_4-class = 2295.238). As shown in Table 2, the information criteria showed only modest decreases between the 3- and 4-class models (ΔAIC ≈ −5; ΔaBIC ≈ −4; BIC slightly increased), suggesting marginal improvement. For context, the BIC is generally more conservative than the AIC, placing a stronger penalty on model complexity.⁵⁴ Additionally, recommendations in the literature⁵⁵ denote that small differences in information criteria (Δ < 10) can justify theory and interpretability in class enumeration, particularly in small samples. The adjusted Lo-Mendell Rubin Likelihood Ratio Test (aLMR LRT) was not significant for any number of classes, but was marginal for the 4 class solution (10.682, p = .073). The entropy value was highest in the 4-class solution (.663), though still only fair. Class sizes for the 4-class solution were adequate for all classes (≥ 16% of the sample). Overall, the fit indices suggested the 4-class solution may be the best fit, but statistical evidence from the class enumeration process alone was not conclusive. Thus, we also consider the theoretical and practical utility of the characteristics of the solutions. As further described in Section 3.1.2 below, the qualitative interpretation of the 4-class solution was consistent with hypotheses based on 1) theories such as Bonanno’s Stress Response and Resilience framework³², 2) modal patterns of PTSS trajectories identified in previous reviews⁵⁶, and 3) PTSS trajectories identified in prior GMMs of very similar samples, including all female samples enriched for CSA exposure.³⁸ The 4-class solution also provided practical, applied utility in clinical settings by distinguishing risk, resilience, recovery, and latent onset of posttraumatic stress symptomology. In summary, all classes in the 4-class solution were well-sized, conceptually coherent, and interpretable, supporting the substantive validity of this model.

Table 2.

Fit indices and entropies for latent growth mixture models.

No. of Classes	AIC	BIC	aBIC	LMR-LRT, p-value	Entropy
1	2289.290	2304.790	2288.960	-	-
2	2254.567	2279.366	2254.038	38.225, p = .205	.553
3	2257.220	2291.318	2256.493	3.141, p = .477	.599
4	2251.840	2295.238	2250.915	10.682, p = .073†	.663
5	2252.887	2305.585	2251.764	4.649, p = .506	.592
6	2254.363	2316.360	2253.041	4.247, p = .218	.623

Note. AIC = Akaike Information Criterion; BIC = Bayesian Information Criterion; aBIC = adjusted Bayesian Criterion; LMR-LRT = Lo-Medell-Rubin likelihood ratio test.

^†p < .08,

^*p <.05,

^**p < .01,

^***p < .001

3.1.2. Characterization of the Final GMM Solution

The final GMM solution (see Figure 1) identified four trajectories of PTSS from late adolescence to mid-life, which we qualitatively describe as: 1) High Stable PTSS; 2) Decreasing PTSS; 3) Increasing PTSS; and 4) Low Stable PTSS. In Class 1, the mean intercept was 13.96 (SD = .59), indicating high PTSS symptoms in adolescence which remained relatively stable from early adulthood to mid-life (M_slopeC1 = −.50, SD = .44). Thus, we describe this trajectory class as High Stable PTSS, comprising 18.0% (n = 29) of the sample (see Figure 1). Class 2, consisting of 16.4% (n = 27) of the sample, had a mean intercept of 10.57 (SD = 1.08), indicating PTSS symptoms above the diagnostic cutoff in adolescence. Class 2 exhibited a negative slope (M_slopeC2 = −2.63, SD = .50), and was thus named Decreasing PTSS. For Class 3, the mean PTSS intercept was 7.78 (SD = .60), just above the diagnostic cutoff for PTSD. The slope in Class 3 was 2.05 (SD = .30), indicating increasing symptoms over time. Thus, we named this class Increasing PTSS, comprising 27.0% (n = 44) of the sample. The mean intercept for Class 4 was 2.69 (SD = .42), indicating low PTSS in late adolescence. The mean slope was 1.10 (SD = .30), indicating relatively stable symptoms from adolescence to mid-life. Thus, we labeled this class Low Stable PTSS, which comprised 38.6% (n = 63) of the sample (see Figure 1). However, the entropy (i.e., separation of classes) in this solution was only fair (.663). We suggest the low entropy across solutions may be related to missing data at some timepoints, thus increasing uncertainty of class assignment. Inspection of the diagonal of the array of Average Latent Class Probabilities for Most Likely Class Membership (see Table S2) indicated that the weak to moderate entropy for the 4-class solution may be driven by variation within the Decreasing PTSS class, and to a lesser extent, the Increasing PTSS class, consistent with the relatively larger SDs around the PTSS intercept and slope observed in these classes. In contrast, other trajectories indicated clearer separation. Any subsequent analyses linking GMM class membership to cortisol functioning used approaches which robustly account for uncertainty in class assignment, partially offsetting concerns about the weak to moderate entropy in the final class solution, though this limitation should still be noted. Finally, child and adolescent PTSD measures were not well validated at the time this study began (T1 = 1987), but T1 sample means and means by PTSS class for depression and dissociation are presented in Table S4 for descriptive purposes.

Figure 1.

PTSS trajectories identified by the 4-class solution

3.2. CSA Exposure and PTSS Class Membership

To apply the R3Step Approach,⁴⁷ we were primarily interested in the High Stable PTSS trajectory as the reference class given prior research that shows CSA may increase risk for chronic PTSS.⁵⁷ As presented in Table 3, CSA status significantly distinguished the High Stable PTSS trajectory from the Low Stable PTSS (p = .004) and Decreasing PTSS trajectories (p = .039), but not the Increasing PTSS class (p = .096). Specifically, participants with CSA exposure were nearly 14 times more likely to belong to the High Stable PTSS class relative to the Low Stable PTSS class (OR = 13.69, 95%CI [2.27, 82.71]) and nearly 13 times more likely to belong to the High Stable relative to the Decreasing PTSS class (OR = 12.97, 95% CI [1.14, 147.02]). Additional analyses using the Low Stable PTSS class as the reference group did not identify any additional significant contrasts (ps > .115). See Table 3 for additional parameterization using all alternate reference classes.

Table 3.

R3Step approach results: Associations between CSA status and PTSS trajectory membership

3. CSA status and PTSS trajectory membership
Class (vs. High Stable)	Estimate (SE)	p-value	OR
Decreasing PTSS	−2.56 (1.24)	.039*	.08
Increasing PTSS	−1.70 (1.02)	.096	.18
Low Stable PTSS	−2.62 (.92)	.004**	.07
Class (vs. Low Stable)	Estimate (SE)	p-value	OR
High Stable PTSS	2.62 (.92)	.004**	13.69
Decreasing PTSS	.05 (.83)	.947	1.06
Increasing PTSS	.91 (.58)	.115	2.49
Class (vs. Increasing)	Estimate (SE)	p-value	OR
High Stable PTSS	1.70 (1.02)	.096	5.49
Decreasing PTSS	−.86 (.96)	.369	.42
Low Stable PTSS	−.91 (.58)	.115	.40
Class (vs. Decreasing)	Estimate (SE)	p-value	OR
High Stable PTSS	2.56 (1.24)	.039*	12.97
Increasing PTSS	.86 (.96)	.369	2.36
Low Stable PTSS	−.05 (.83)	.947	.95

Note. Estimates are unstandardized B from latent variable multinomial logistic regressions using the 3-step procedure (R3Step).

^*p < .05,

^**p < .01,

^***p < .001

3.3.1. Associations Between PTSS Class Membership and HPA Attenuation

Cortisol Intercept.

We next examined associations between PTSS trajectory membership and individual cortisol slopes and intercepts extracted from the MLM using the R3Step approach. Cortisol intercept reflects basal serum morning cortisol levels at T1, corresponding to the acute stage of trauma for CSA-exposed participants. The average cortisol intercept for the full sample was 7.38 (SD = .97). As presented in Table 4a, cortisol intercept significantly distinguished the High Stable PTSS and Low Stable PTSS trajectories, such that for every one unit increase in cortisol intercept concentration (i.e., increase by 1.0 μg/dl), the likelihood of belonging to the High Stable PTSS trajectory relative to the Low Stable PTSS trajectory nearly tripled (OR = 2.94, 95% CI [1.45, 5.94], p = .003). Cortisol intercept also distinguished the High Stable PTSS trajectory from the Increasing PTSS trajectory, such that for each 1-unit increase in cortisol intercept, the likelihood of belonging to the High Stable PTSS trajectory relative to the Increasing PTSS trajectory increased by 172% (OR = 2.72, 95% CI [1.15, 6.42], p = .023). Finally, cortisol intercept distinguished the Decreasing PTSS and Low Stable PTSS trajectories, such that for every 1-unit increase in cortisol intercept, the likelihood of belonging to the Decreasing PTSS trajectory doubled relative to the Low Stable PTSS trajectory (OR = 2.00, 95% CI [1.03, 3.87], p = .041). For all contrasts, including non-significant trajectory contrasts, see Table 4a. Mean cortisol intercepts by PTSS trajectory membership (calculated via BCH) are also presented in Table 4a for descriptive purposes. The overall pattern of results appears to indicate that higher cortisol levels at T1 in early adolescence (M_age = 11.03) is linked with membership in PTSS trajectories characterized by higher symptomology at T4 (late adolescence, M_age = 18.06), when PTSS trajectory estimates begin.

Table 4.

Associations between PTSS trajectory membership and HPA Outcomes

4a. Cortisol Intercept					4b. Cortisol Slope
R3Step Output				BC H	R3Step Output					BC H
Class (vs. High Stable)	Estimate (SE)	p	OR	Mean	Class (vs. High Stable)	Estimate (SE)	p	OR (raw)	OR (std)	Mean
Decreasing PTSS	−.39 (.37)	.292	.68	7.73	Decreasing PTSS	2.22 (3.91)	.570	9.17	1.27	.26
Increasing PTSS	−1.00 (.44)	.023*	.37	7.14	Increasing PTSS	10.13 (4.33)	.019*	25002.78	3.05	.35
Low Stable PTSS	−1.08 (.36)	.003**	.34	7.10	Low Stable PTSS	9.90 (3.49)	.005**	19918.39	2.97	.35
Class (vs. Increasing)	Estimate (SE)	p	OR	Mean	Class (vs. Increasing)	Estimate (SE)	p	OR (raw)	OR (std)	Mean
High Stable PTSS	1.00 (.44)	.023*	2.72	8.04	High Stable PTSS	−10.13 (4.33)	.019*	.00	.33	.25
Decreasing PTSS	.61 (.41)	.132	1.84	7.73	Decreasing PTSS	−7.91 (4.21)	.060†	.00	.42	.26
Low Stable PTSS	−.08 (.33)	.814	.92	7.10	Low Stable PTSS	−.23 (2.70)	.933	.80	.98	.35
Class (vs. Low Stable)	Estimate (SE)	p	OR	Mean	Class (vs. Low Stable)	Estimate (SE)	p	OR (raw)	OR (std)	Mean
High Stable PTSS	1.08 (.36)	.003**	2.94	8.04	High Stable PTSS	−9.90 (3.49)	.005**	.00	.34	.25
Decreasing PTSS	.69 (.34)	.041*	2.00	7.73	Decreasing PTSS	−7.68 (3.41)	.024*	.00	.43	.26
Increasing PTSS	.08 (.33)	.814	1.08	7.14	Increasing PTSS	.23 (2.70)	.933	1.26	1.03	.35
Class (vs. Decreasing)	Estimate (SE)	p	OR	Mean	Class (vs. Decreasing)	Estimate (SE)	p	OR (raw)	OR (std)	Mean
High Stable PTSS	.39 (.37)	.292	1.47	8.04	High Stable PTSS	−2.22 (3.91)	.570	.11	.79	.25
Increasing PTSS	−.61 (.41)	.132	.54	7.14	Increasing PTSS	7.91 (4.21)	.060†	2727.66	2.39	.35
Low Stable PTSS	−.69 (.34)	.041*	.50	7.10	Low Stable PTSS	7.68 (3.41)	.024*	2172.98	2.33	.35
4c. Cortisol Intercept (Residualized for Covariates)					4d. Cortisol Slope (Residualized for Covariates)
R3Step Output				BC H	R3Step Output					BC H
Class (vs. High Stable)	Estima te (SE)	p	OR	Mea n	Class (vs. High Stable)	Estima te (SE)	P	OR (raw)	OR (st d)	Mea n
Decreasing PTSS	.54 (.38)	.156	.67	.33	Decreasing PTSS	2.30 (3.79)	.543	10.02	1.29	−.048
Increasing PTSS	.94 (.41)	.022*	.39	−.21	Increasing PTSS	9.24 (3.92)	.019*	10299.70	2.76	.030
Low Stable PTSS	−.12 (.32)	.003**	.35	−.27	Low Stable PTSS	9.81 (3.31)	.003**	18115.59	2.94	.034
Class (vs. Increasing)	Estimate (SE)	p	OR	Mean	Class (vs. Increasing)	Estimate (SE)	p	OR (raw)	OR (std)	Mean
High Stable PTSS	.94 (.41)	.022*	2.55	.66	High Stable PTSS	−9.24 (3.92)	.019*	.00	.36	−.067
Decreasing PTSS	.54 (.38)	.156	1.72	.33	Decreasing PTSS	−6.94 (3.99)	.082	.00	.47	−.048
Low Stable PTSS	−.12 (.32)	.713	.89	−.27	Low Stable PTSS	.57 (2.61)	.829	1.76	1.06	.034
Class (vs. Low Stable)	Estimate (SE)	p	OR	Mean	Class (vs. Low Stable)	Estimate (SE)	p	OR (raw)	OR (std)	Mean
High Stable PTSS	1.06 (.35)	.003**	2.88	.66	High Stable PTSS	−9.81 (3.31)	.003**	.00	.34	−.067
Decreasing PTSS	.66 (.33)	.047*	1.94	.33	Decreasing PTSS	−7.50 (3.39)	.027*	.00	.44	−.048
Increasing PTSS	.12 (.32)	.713	1.13	−.21	Increasing PTSS	−.57 (2.61)	.829	.57	.94	.030
Class (vs. Decreasing)	Estimate (SE)	p	OR	Mean	Class (vs. Decreasing)	Estimate (SE)	p	OR (raw)	OR (std)	Mean
High Stable PTSS	.39 (.37)	.283	1.48	.66	High Stable PTSS	−2.30 (3.79)	.543	.10	.78	−.067
Increasing PTSS	−.54 (.38)	.156	.58	−.21	Increasing PTSS	6.94 (3.99)	.082	1028.33	2.14	.030
Low Stable PTSS	−.66 (.33)	.047*	.52	−.27	Low Stable PTSS	7.50 (3.40)	.027*	1808.67	2.28	.034

Note. Raw odds ratios computed for the effect of cortisol slope on PTSS trajectory membership reflect the effect of a one-unit increase in cortisol slope on the likelihood of belonging to one PTSS trajectory relative to another. However, the mean cortisol slope in this sample was .32, with a range of −.12 to .62. Thus, a one unit increase in slope is larger than the entire range of cortisol slope values, leading to very large (and very small) odds ratios that are difficult to interpret. Raw odds ratios were thus supplemented with computation of odds ratios for standardized cortisol slope scores, where the “OR (std)” reflects the effect of a one standard deviation unit increase in cortisol slope on the likelihood of belonging to one PTSS class relative to another. These OR (std) values are the results reported in-text.

^†p < .08,

^*p < .05,

^**p < .01,

^***p < .001.

SE = standard error, OR = odds ratio.

Cortisol Slope.

Cortisol slope analyses were similarly conducted via the R3Step approach, see Table 4b. Consistent with previous research,¹³ lower (i.e., less steep) cortisol slopes were theorized to indicate HPA attenuation. The average cortisol slope for the full sample was .32 (SD = .11), indicating modestly increasing morning basal cortisol levels with age on average. As presented in Table 4b, cortisol slope scores significantly distinguished the Low Stable PTSS and High Stable PTSS trajectory, such that for each 1-standard deviation increase in cortisol slope (reflecting less attenuation), the likelihood of belonging to the Low Stable PTSS trajectory relative to the High Stable PTSS trajectory nearly tripled (OR = 2.97, 95% CI [1.40, 6.31], p = .005). Cortisol slope also distinguished the Low Stable PTSS trajectory and Decreasing PTSS trajectory, such that for each 1-standard deviation increase in cortisol slope, the likelihood of belonging to the Low Stable PTSS trajectory relative to the Decreasing PTSS trajectory more than doubled (OR = 2.33, 95% CI [1.12, 4.88], p = .024). Cortisol slope further distinguished the Increasing PTSS and High Stable PTSS trajectories, such that for each 1-standard deviation unit increase in cortisol slope, the likelihood of belonging to the Increasing PTSS class relative to the High Stable PTSS class was approximately triple (OR = 3.05, 95% CI [1.20, 7.77], p = .019). Finally, the effect of cortisol slope on membership in the Increasing PTSS vs. Decreasing PTSS trajectories was marginal, such that each 1-standard deviation unit increase in cortisol slope was associated with 2.4 times higher likelihood of belonging to the Increasing PTSS trajectory relative to the Decreasing PTSS trajectory (OR = 2.39, 95% CI [.960, 5.95], p = .060). Mean cortisol slopes by PTSS trajectory membership calculated via the BCH approach are also presented in Table 4b for descriptive purposes. The overall pattern of results appears to indicate that less steep cortisol slopes from early adolescence (M_age = 11.09) to early adulthood (M_age = 24.4), reflecting greater HPA attenuation over this period, were linked to increased odds of belonging to trajectories characterized by higher symptomology at T4 (late adolescence, M_age = 18.06), when PTSS trajectory estimates begin, namely the High Stable PTSS and Decreasing PTSS trajectories.

3.3.2. Sensitivity Analysis

Per the analytic plan, we evaluated whether associations between PTSS trajectory membership and HPA functioning were robust to demographic covariates (i.e., maternal education, participant racial minority status) by rerunning R3Step analyses on residualized cortisol intercept and slope values. As presented in Table 4c, cortisol intercept scores residualized for covariates significantly distinguished the High Stable PTSS and Low Stable PTSS trajectories, such that for every one unit increase in residualized morning cortisol intercept score, the likelihood of belonging to the High Stable PTSS trajectory relative to the Low Stable PTSS trajectory nearly tripled (OR = 2.99, 95% CI [1.44, 5.73], p = .003). Residualized cortisol intercept scores also distinguished the High Stable PTSS trajectory from the Increasing PTSS trajectory, such that for each 1-unit increase in residualized cortisol intercept score, the likelihood of belonging to the High Stable PTSS trajectory relative to the Increasing PTSS trajectory increased by 155% (OR = 2.55, 95% CI [1.15, 5.69], p = .022). Finally, cortisol intercept distinguished the Decreasing PTSS and Low Stable PTSS trajectories, such that for every 1-unit increase in residualized cortisol intercept score, the likelihood of belonging to the Decreasing PTSS trajectory approximately doubled relative to the Low Stable PTSS trajectory (OR = 1.94, 95% CI [1.01, 3.73], p = .047). Turning to residualized cortisol slope scores (see Table 4d), results indicated that cortisol slope scores residualized for demographic covariates continued to significantly distinguish the Low Stable PTSS and High Stable PTSS trajectories, such that for each 1-unit increase in residualized cortisol slope score (reflecting less attenuation), the likelihood of belonging to the Low Stable PTSS trajectory relative to the High Stable PTSS trajectory nearly tripled (OR = 2.94, 95% CI [1.44, 5.99], p = .003). Residualized cortisol slope scores also distinguished the Low Stable PTSS trajectory and Decreasing PTSS trajectory, such that for each 1-standard deviation increase in residualized cortisol slope score, the likelihood of belonging to the Low Stable PTSS trajectory relative to the Decreasing PTSS trajectory more than doubled (OR = 2.28, 95% CI [1.10, 4.73], p = .027). After residualizing for covariates, cortisol slope further distinguished the Increasing PTSS and High Stable PTSS trajectories, such that for each 1-standard deviation unit increase in residualized cortisol slope score, the likelihood of belonging to the Increasing PTSS class relative to the High Stable PTSS class nearly tripled (OR = 2.76, 95% CI [1.18, 6.42], p = .019). However, after accounting for covariates, the effect of residualized cortisol slope scores on membership in the Increasing PTSS vs. Decreasing PTSS trajectories was non-significant (OR = 2.14, 95% CI [.91, 5.06], p = .082). Descriptive means for each PTSS trajectory for cortisol slopes and intercepts calculated via the BCH approach are also presented in Table 4c and 4d. Overall, sensitivity analyses for both residualized morning cortisol slope and intercept values generally affirm primary results.

3.3.3. Post-Hoc Analyses: The Role of Revictimization

Results above indicating that steeper cortisol slopes (reflecting less HPA attenuation) were linked to membership in the Increasing PTSS trajectory were somewhat surprising. Given subsequent trauma exposure later in development may increase PTSS,⁵³ we further contextualized our findings by evaluating associations between non-sexual revictimization (see Table 5a) and trajectory membership. For every additional type of non-sexual traumatic event reported, participants exhibited 27% greater likelihood of belonging to the Increasing PTSS Class relative to the Low Stable PTSS class (OR = 1.27, 95%CI [1.02, 1.59], p = .032) and a 31% greater likelihood of belonging to the Increasing PTSS Class relative to the Decreasing PTSS class (OR = 1.31, 95% CI[1.01, 1.68], p = .040). For each additional traumatic event reported, participants also had a 24% greater likelihood of belonging to the High Stable PTSS class relative to the Low Stable PTSS class (OR = 1.24, 95%CI [1.01, 1.52], p = .041) and a 27% greater likelihood of belonging to the High Stable PTSS class relative to the Decreasing PTSS class, a marginally significant difference (OR = 1.27, 95% CI [.98, 1.64], p = .072). See Table 5a for additional parameterization using all alternate reference classes. Presented descriptively (i.e., via BCH approach), the High PTSS class reported the most non-sexual PTEs on average at 10.93, followed by the Increasing PTSS class at 10.31. The Decreasing PTSS and Low Stable PTSS classes reported comparatively fewer non-sexual PTEs at 5.45 and 6.27 on average respectively.

Table 5.

Post-hoc analyses of associations between sexual and non-sexual revictimization and PTSS trajectory membership.

5a. Non-Sexual PTEs and PTSS Trajectory Membership				5b. Adult Sexual Victimization and PTSS Trajectory Membership
Class (vs. High Stable)	Estimate (SE)	p-value	OR	Class (vs. High Stable)	Estimate (SE)	p-value	OR
Decreasing PTSS	−.24 (.13)	.072†	.79	Decreasing PTSSa	−54.4 (.00)	<.001***	.00
Increasing PTSS	.03 (.07)	.679	1.03	Increasing PTSS	.43 (.86)	.619	1.53
Low Stable PTSS	−.21 (.10)	.041*	.81	Low Stable PTSS	−1.19 (.70)	.089	.31
Class (vs. Increasing)	Estimate (SE)	p-value	OR	Class (vs. Increasing)	Estimate (SE)	p-value	OR
High Stable PTSS	−.03 (.07)	.679	.97	High Stable PTSS	−.43 (.86)	.619	.65
Decreasing PTSS	−.27 (.13)	.040*	.77	Decreasing PTSSa	−54.83 (.86)	<.001***	.00
Low Stable PTSS	−.24 (.11)	.032*	.79	Low Stable PTSS	−1.62 (.72)	.025*	.20
Class (vs. Low Stable)	Estimate (SE)	p-value	OR	Class (vs. Low Stable)	Estimate (SE)	p-value	OR
High Stable PTSS	.21 (.10)	.041*	1.24	High Stable PTSS	1.19 (.70)	.089	3.28
Decreasing PTSS	−.03 (.13)	.842	.98	Decreasing PTSSa	−53.21 (.70)	< .001***	.00
Increasing PTSS	.24 (.11)	.032*	1.27	Increasing PTSS	1.62 (.72)	.025*	5.03
Class (vs. Decreasing)	Estimate (SE)	p-value	OR	Contrasts with Decreasing PTSS trajectory are not reported due to excessively large, potentially unreliable OR estimates, as no participants in the Decreasing PTSS trajectory endorsed adult sexual victimization in this sample.
High Stable PTSS	.24 (.13)	.072†
Increasing PTSS	.27 (.13)	.040*	1.31
Low Stable PTSS	.03 (.13)	.842	1.03

Note.

^aNone of the participants in the Decreasing PTSS trajectory endorsed adult sexual victimization, resulting in difficulties producing reliable Odds Ratio estimates and should be interpreted cautiously.

^†p < .08,

^*p < .05,

^**p < .01,

^***p < .001.

PTEs = potentially traumatic events. SE = standard error, OR = odds ratio.

Results from R3Step analyses indicated that participants reporting adult sexual victimization were five times more likely to belong to the Increasing PTSS class relative to the Low Stable PTSS class (OR = 5.03, B = 1.62, SE = .72, p = .025, see Table 5b). Participants reporting adult sexual victimization were also less likely to belong to the Decreasing PTSS trajectory relative to all other trajectory classes (ps < .001). See Table 5b for additional contrasts, although no other significant contrasts emerged (ps > .089). Overall, results indicate adult sexual revictimization increases the likelihood of belonging to a trajectory experiencing increasing PTSS symptoms in adulthood. Presented descriptively (i.e., via BCH approach), 23.4% of the Low Stable PTSS trajectory endorsed adult sexual victimization, compared to 50.6% in the High Stable PTSS trajectory and 61.2% in the Increasing PTSS trajectory. None of the participants assigned to the Decreasing PTSS class endorsed adult sexual victimization, resulting in Odds Ratios of 0.00 and difficulties producing Confidence Intervals for these estimates, suggesting contrasts with the Decreasing PTSS trajectory should be interpreted cautiously.

4.0. Discussion

Results from this longitudinal, prospective study identified four person-centered trajectories of PTSS from late adolescence to mid-life: High Stable, Increasing, Decreasing, and Low Stable PTSS. This constitutes one of the longest spanning GMMs of PTSSs over time generally, and the single longest spanning evaluation of PTSS trajectories ever conducted in a maltreated sample. In addition to these descriptive contributions, the current study expands on previous research conducted with this same cohort by Trickett and colleagues,¹³ which found that CSA was linked to HPA attenuation, as indicated by significantly higher morning cortisol in the period after initial disclosure of CSA and a significantly less steep slope of morning cortisol over development. Building on this earlier work, the current findings further clarify how CSA and HPA attenuation are implicated in the long-term course of PTSS symptoms. Consistent with previous research in other samples,⁵⁷ we find CSA exposure is associated with higher likelihood of belonging to the High Stable PTSS relative to other trajectories, namely the Low Stable PTSS and Decreasing PTSS trajectories. This underscores the pernicious long-term effects of CSA, with results signaling CSA elevates risk for PTSS into midlife.

The most notable contributions of the current study, however, concern HPA attenuation. Sensitivity analyses further indicated HPA attenuation findings were generally robust to demographic confounders. Specifically, we evaluated HPA attenuation using inter-individual differences in intra-individual change by estimating morning cortisol intercepts and slopes over six timepoints from early adolescence to early adulthood, where high initial morning cortisol levels and less steep cortisol slopes from T1 to T6 (corresponding approximately to ages 11 to 24, i.e., early adolescence to early adulthood) were theorized to represent greater HPA attenuation. Results show developmental HPA attenuation indexed in this manner was significantly associated with the course of PTSS from late adolescence to midlife. We found that individuals in the High Stable PTSS class exhibited the strongest evidence of HPA attenuation over development, characterized by higher morning cortisol levels in early adolescence and less steep increases in morning cortisol from early adolescence to early adulthood, relative to the Low Stable and Increasing PTSS classes. A similar, though less pronounced, pattern was observed for individuals in the Decreasing PTSS class, who also exhibited higher initial cortisol levels and flatter cortisol slopes compared to the Low Stable PTSS class. We interpret this to indicate that individuals in the Decreasing PTSS trajectory also exhibited HPA attenuation that may be understood as an intermediate biological embedding pattern, which may drive initial stress reactivity and early symptom elevation rather than determining long-term PTSS course. Thus, individuals who exhibit chronic symptoms versus a subsequent recovery may be further differentiated by other factors, including presence of comorbidities, social support, and treatment access.^58,59 It should also be noted that PTSS trajectories were estimated beginning at T4, approximately 6 years after study enrollment, thus members of the Decreasing PTSS class still had fairly high symptomology several years after initial CSA exposure and did not recover until adulthood.

Based on the current findings, we cautiously suggest that HPA attenuation may not only serve as a biomarker of risk but could actively influence individuals’ lived experiences of trauma and trauma-related symptoms. Attenuated cortisol patterns may contribute to heightened threat sensitivity, difficulties with stress regulation, and prolonged emotional arousal, which in turn reinforce chronic PTSS trajectories. Conversely, less pronounced HPA attenuation may support more flexible stress responding and facilitate recovery. This dual role underscores the potential value of targeting stress physiology in early interventions, while also highlighting cortisol as an informative marker for identifying individuals at elevated risk for persistent PTSS. Overall, the current findings may serve to clarify the pathophysiology of PTSS symptom course by characterizing the role of developmental HPA attenuation processes in shaping long-term PTSS trajectories.

Longitudinal biological investigations of PTSD are rare, but much needed. By indexing biological embedding as a developmental process using longitudinally-derived cortisol slopes and intercepts, the current design aims to accurately capture the gradual, developmental nature of HPA attenuation processes.^5,13,24 Limited prior longitudinal work examining PTSS course and cortisol examine symptom course over a much shorter time range, typically 2 years or less.⁶⁰ This complicates comparison to the current results, which focus on characterizing the role of long-term HPA attenuation. Nonetheless, our findings are generally consistent with core hypotheses from biological embedding frameworks that emphasize that the consequences of early chronic stress are most evident in within-person changes in stress physiology across development, rather than isolated assessments of cortisol levels.^11,25,26 These models propose that repeated activation of the HPA axis in response to sustained environmental stress produces cumulative allostatic overload, which over long spans of time may result in compensatory downregulation of cortisol, often conceptualized as HPA attenuation.^13,24,61,62 Prior research has demonstrated that basal HPA-axis function in youth is associated with both current and future emotional and behavioral outcomes,^6,63 yet few studies have examined how developmental patterns of cortisol relate to heterogeneous symptom courses following trauma exposure. In doing so, the current findings extend prior work limited to narrow developmental windows (e.g., adult cross-sectional studies of PTSD neuroendocrine correlates^64,65) and provide empirical work for models⁶⁶ implicating stress physiology in the maintenance of PTSS over time. As applied to associations with posttraumatic stress symptomology course, persistent PTSSs may both reflect and reinforce this recalibration of the HPA axis through sustained threat appraisal and corticotropin-releasing hormone drive, contributing to attenuated cortisol slopes observed years after exposure.^67–70 Thus, while attenuated cortisol may serve as a biomarker of risk, it may also reflect developmental recalibration of stress physiology that contributes to sustained hypervigilance or altered threat appraisal, influencing the persistence of PTSS over time. Understanding these patterns can inform personalized approaches to monitoring and intervention for individuals at risk for chronic PTSS.

Post-hoc analyses examined the role of subsequent potentially traumatic events in distinguishing posttraumatic stress trajectories. For each additional non-sexual traumatic event endorsed, participants were approximately 30% more likely to belong to the High Stable or Increasing PTSS classes over the Low Stable or Decreasing PTSS Classes. Adult sexual trauma was a particularly potent predictor of class membership, with adult sexual revictimization conferring nearly three times higher likelihood of belonging to the High Stable PTSS class over the Low Stable PTSS class and over five times higher odds of belonging to the Increasing PTSS class over the Low Stable PTSS class. Thus, revictimization may interfere with PTSS recovery, consistent with previous research.⁷¹ These findings also point to etiological differences for differing PTSS trajectories. For example, the Increasing PTSS class exhibited HPA functioning patterns more similar to the Low Stable PTSS class, with current results suggesting their increasing PTSS symptoms may be driven by subsequent trauma experienced in adulthood, particularly adult sexual assault, rather than HPA attenuation. These differences in PTSS etiology have clinical implications, as members of the High PTSS and Increasing PTSS trajectories may have similar clinical presentations at midlife, but require differentiated treatment based on differing etiology.

Although not directly tested here, we put forth some tentative hypotheses for clinical conceptualizations based on the current results to guide future intervention research. While there are likely common treatment mechanisms that will benefit multiple PTSD presentations, like trauma-informed exposure techniques, we suggest there are also nuanced considerations based on symptom etiology. Current results point to a subgroup of CSA-exposed individuals with clinical presentations characterized by 1) early and persistent biological embedding in the HPA system, 2) more entrenched PTSSs such as threat bias and avoidance, and 3) more recursive biological embedding, where chronically high trauma symptoms may reinforce HPA attenuation via a feedback loop. This presentation implies individuals with symptom trajectories similar to the High Stable PTSS group identified in this study may benefit from a multi-modal, gradual treatment approach where slow paced exposure therapy is paired with pauses for physiological regulation (e.g., biofeedback,⁷² mindfulness^73,74). Empirical research on the effects of behavioral interventions on physiological stress recalibration is limited, but emerging research suggests exposure therapy⁷⁵ and cognitive behavioral therapy⁷⁶ may target HPA functioning. Additionally, the Increasing PTSS trajectory identified in the current results is consistent with prior conceptualizations of delayed onset PTSD, with common presentations including subclinical symptoms with later increases in symptoms precipitated by later life stressors.⁷⁷ This symptom trajectory is consistent with stress sensitization and kindling models, whereby repeated stress exposures progressively activate neurobiological systems, including fear circuits and the HPA axis, eventually overwhelming coping abilities.⁷⁷ These individuals may benefit from dedicated space to discuss and process the particular stressor that precipitated onset of their delayed symptoms. Additionally, this delayed onset group demonstrated initial resilience, suggesting they may already have some adaptive coping skills that can be re-accessed and reinforced in the therapeutic context.

Based on these results, we also advance policy recommendations that can prevent CSA and violence against women. Examples of such legislation from the US-context include Erin’s Law, which mandates CSA prevention education in schools. Expanding the number of locales implementing similar laws and increasing funding to better support efficacious primary prevention of CSA can reduce the personal, societal, and medical system burden of CSA. Concerning the identified role of revictimization, the Violence Against Women Act aims to support women affected by domestic violence and sexual assault by providing trauma-informed training for law enforcement, prosecutors, and judges, enhancing the enforceability of protective orders, and providing housing protection for survivors of domestic and sexual violence. The current version of the Violence Against Women Act is set to expire in 2027, underscoring the importance of continued support for this legislation.

From a scientific perspective, the novelty of the current findings is bolstered by methodological strengths, including a longitudinal, cross-sequential design spanning nearly 30 years. A deliberate recruitment strategy yielded a trauma-enriched sample where approximately half of participants were exposed to CSA and timing of CSA exposure was closely tied to baseline data. Moreover, comparison participants without CSA exposure were rigorously matched on demographic criteria. We view the focus on CSA as a potentially traumatic event as a strength, as CSA is associated with higher rates of PTSD compared to other potentially traumatic events,^27,28 which resulted in sufficient base rates and variability in PTSS course for the current investigation. While an all-female sample reduces generalizability, prior research shows women develop PTSD at higher rates than men.²⁹ Thus, the focus on female participants parallels populations of interest in the study of PTSD. Furthermore, the use of morning cortisol intercept and slope parameters to index interindividual differences in intraindividual HPA attenuation constitutes a sophisticated, intensive analytic approach, with sensitivity analyses confirming findings are robust to demographic confounders. Finally, post-hoc analyses contextualize the role of sexual and non-sexual revictimization, which highlight how the course of PTSS is not only related to physiology, but also life course history.

However, these contributions are tempered by several limitations worth noting. First, the current results are based on a relatively small sample, although the small sample size is somewhat offset by the merits of the longitudinal, prospective design which spans approximately 30 years with high retention. That being said, there is partial overlap in the timing of assessment of cortisol (T1-T6, average ages 11–24) and PTSS (T4-T7, 18–37). This is consistent with a biological embedding framing, whereby HPA attenuation may primarily drive PTSS symptom course, but also simultaneously be subject to dynamic, potentially bidirectional transactions between stress physiology and psychological processes over time.²⁵ The GMM class enumeration process pointed to the 4-class solution, but several aspects of the fit indices across the model enumeration process suggest caution in interpreting these trajectories as definitively representative of latent PTSS trajectories. Fit indices, including a trend-level aLMR-LRT, modest improvements in AIC and aBIC across solutions, and somewhat weak entropy indicate statistical evidence alone was not conclusive. Thus, we relied in part on justification according to theory and prior empirical studies of PTSS trajectories, which may introduce bias by privileging theoretically expected patterns. Although these statistical limitations introduce the possibility that fewer or different latent trajectories could exist in this sample, the 4-class solution aligns with prior theoretical frameworks and empirical research on trauma-response trajectories in similar populations, including recovery, resilience, chronic, and delayed-onset patterns, and provides conceptually coherent, clinically meaningful categories. Nevertheless, these classes should be interpreted as provisional and exploratory, and conclusions regarding PTSS heterogeneity should be tempered by the limitations of the data and model fit. Furthermore, the entropy of the final 4-trajectory solution to the GMM indicated weak to moderate class separation between PTSS trajectories. Further inspection of results indicated there may be some “fuzziness” or within-class variability in the Decreasing PTSS class in particular, suggesting that the pace and process of recovering from PTSS may vary from person to person. Concerns regarding this somewhat low entropy are partially offset by the application of the R3Step approach for focal analyses, which accounts for uncertainties in class membership using weighting procedures, but should still be interpreted with this limitation in mind.⁵² While the PTSS trajectories identified in the current study spanned over a decade, the PTSS trajectories were based on only 3 timepoints and we only modeled linear trajectories to avoid overfitting. Thus, there may have been nonlinear symptom fluctuation between timepoints not captured in the current investigation, which may also have contributed to the relatively low entropy in the final class solution.

Finally, indexing cortisol in the context of a long-ranging longitudinal study is an ambitious endeavor, with concomitant technological developments and changes in recommended best practices for assessment. While the methods for converting serum and saliva concentrations of cortisol follow precedent from previously published studies,¹³ we cannot guarantee these methodological changes over the course of the study did not impact results. For example, serum saliva was collected via indwelling catheter (i.e., “blood draw”), which may be experienced as stressful by some participants, particularly in comparison to less invasive saliva sampling methods. Additionally, while the serum to saliva conversion formula used in this study has been used in previous publications,^13,51 some methodologists⁷⁸ argue that salivary cortisol may more closely approximate biologically active cortisol, because salivary cortisol measures only unbound cortisol, whereas serum cortisol includes unbound cortisol and cortisol bound to receptors, suggesting later measurements in this study may more accurately index biologically active cortisol levels. Finally, the serum-saliva conversion formula used here was derived from a different sample, thus mismatch in major characteristics of the conversion formula validation sample and the current sample could potentially bias results due to inaccurate coefficients. Likewise, timing of cortisol sampling was standardized within 1-hour (9:00–10:00am) based on contemporary guidance when study protocols were designed, but may still be affected by diurnal rhythms. Study protocols did not include assessments of waking time, which may further confound results. Additionally, this study assessed morning basal cortisol specifically, which approximates peak diurnal activation, and thus does not fully capture generalized HPA axis function nor do the current data reflect total cortisol output. Nevertheless, peak or early-morning cortisol may be particularly informative in traumatized samples, as alterations in peak activation are thought to reflect shifts in HPA set-point or capacity for stress-system engagement following chronic early adversity. Finally, we indexed HPA attenuation by modeling linear cortisol slopes and intercepts based on precedent from prior empirical investigations with this cohort.^13,51 However, we cannot rule out the possibility that curvilinear patterns of long-term developmental HPA changes may exist. While few (if any) other existing cohorts have cortisol data spanning developmental periods as long as those examined here, this would be a meaningful future direction, particularly in larger samples with greater statistical power to detect nonlinear associations. That said, the pattern of linear HPA attenuation observed here may be more accurately understood as CSA-related HPA attenuation, as it is possible other forms of trauma or other cohorts may exhibit distinct potentially non-linear HPA trajectories.

Despite these limitations, the current study advances the literature by identifying PTSS trajectories spanning over a decade, the longest period evaluated in a maltreated sample to date while simultaneously providing empirical support for the theorized association between HPA attenuation and PTSS symptom course. We further suggest that HPA attenuation measures may provide both predictive and mechanistic information beyond trauma exposure or symptom levels alone, offering insight into why individuals with similar trauma histories may exhibit divergent symptom trajectories, consistent with the concept of multifinality. Such information may ultimately support early identification of individuals at the greatest risk for chronic PTSS, guiding targeted intervention and informing public health strategies aimed at mitigating long-term sequelae of trauma. Finally, these results point to potential future directions in research. With regard to research implications, we recommend future studies characterize the role of timing and duration of trauma exposure in shaping biological embedding processes, such as through the application of structured life course modeling.⁷⁹ Such approaches may be able to more precisely disentangle timing and directionality nuances of the potential bidirectional interactions between biological and psychological stress processes, in which HPA attenuation may influence emotional and cognitive responses to stress, while persistent symptoms such as hypervigilance and intrusive memories may further reinforce HPA attenuation. Moreover, future research should consider how PTSS trajectories may be related to other aspects of stress response coordination, including cortisol reactivity, diurnal cortisol rhythms, and coordination with other stress response system components, particularly the sympathetic nervous system. Overall, results further underscore both the long-term, pernicious effects of CSA exposure and the role of revictimization in initiating PTSS and sustaining chronically high PTSS, highlighting the need for urgent public health intervention.

Highlights:

• Posttraumatic stress shows chronic, recovering, resilient, and delayed courses

• HPA attenuation is linked with chronic posttraumatic stress trajectories

• Results imply trauma treatment should be tailored to symptom and biological history